Machine Learning Techniques

In today's world, we are surrounded by machines and different types of gadgets. Machine learning is a technique which prepares machines to think and learn. In modern days machine learning can help human a lot. In machine learning what we do is to analyse a set of data to get some predictions. We can use machine learning in many fields like medical, agriculture, climate, engineering, and many other domains. In this article, we are going to explore some techniques which are used to analyse a set of data. This technique can be divided into two parts as follows:

1. Supervised learning: In this type of machine learning, we use a set of data to predict. In this process, a model is created to analyse set of input data and set of output data. This model takes input and output data which are provided and verified. After that it gives us reasonable and logical predictions based on given data. This method can only be used when we have response means output for input data.
2. Unsupervised learning: When we have no response means output data for input set then we have to use unsupervised learning. It mainly finds out the internal pattern of the model and then predicts the output for new data.

Difference between supervised and unsupervised learning

Supervised learning	Unsupervised learning
1. It uses input and output data.	1. It does not use responses.
2. It cannot work without responses.	2. It can work without responses.
3. It needs supervision to do work.	3. It does not need any supervision.
4. In this method, both input and output data are provided to the model.	4. In this method, only input data is provided to the model.
5. Regression and classification are supervised learning.	5. Clustering and association are unsupervised learning.
6. Supervised model predicts the output.	6. Unsupervised model finds the internal pattern.
7. Supervised learning model can be used in the cases where we are provided both input and output.	7. Unsupervised learning model can be used in the cases where we are provided only input data set.

Following is a list of ten methods that you must know to improve your machine learning knowledge and skills:

• Regression
• Classification
• Clustering
• Dimensionality Reduction
• Ensemble Methods
• Neural Nets and Deep Learning
• Transfer Learning
• Reinforcement Learning
• Natural Language Processing
• Word Embedding's

1. Regression:  In this method, we can predict a particular numerical value by checking previous input and output sets. Now, it is clear that it is supervised machine learning. In this method there are simple to complex approaches. Linear regression is the simplest approach. In this approach, we use a simple and easy equation of the straight line which is y = mx + c. We try to find out the slope for different (x,y).
   
   Example: As an example we can say about the prediction of energy consumption of a building when other factors of the building are provided. We can find our answer from the slope which we get from other building’s information.
   
2. Classification: In this method, we can classify one new input based on the input and output data provided to the model. It is also supervised machine learning. There exists many models for classification and the simplest one is logistic regression. Don’t confuse with the word regression. It is not a regression method. In this model the probability of occurrence is checked based on the previous input and output datasets. There are also other models like decision trees, support vector machines, random forests and others.
   
   Example: As an example, we can say about checking a student’s probability to get admission in a college. We check student will get admission or not, it means the answer will be “yes or no”. In same pattern, we can predict a buyer will buy a product or not. In logistic regression method, we get the probability which is in numerical form. The probability can have the value from 0 to 1. If we get the probability of a student’s admission near about 1 then we can say that he will get admission and if we get probability near about 0 then we can say the student will not get the admission. If the probability is 0.5 then we take it as no.
   
3. Clustering: It is an untrained method of machine learning. It does not use any output data set but the model gives us the output for a new input. Here, we can use only visualization to observe the solution. K- Means method is most popular method of clustering. Here K represents the number of clusters, which are chosen by the user. In this method, we will choose K random centres among the data set. Then we put the data points in such a manner that they are closest to the random data centres. The process is over if the centres do not change (or change very little).
   
   Example: We can use the previously discussed building’s electric consumption problem. We can use other building’s properties to describe it.
   
4. Dimensionality Reduction: Suppose, you have a big data set. Now to analyze this data, you need a large time. If there exists some information which is not relevant or you can identify them as not much more important then you will want to reduce them from your data set. In this case, you require dimensionality reduction. In this method, you can efficiently cut the extra information from your data set. There are many methods for dimensionality reduction. The most popular method is PCA (Principal Complexity Analysis). It reduces the dimension by finding new vector which maximize the linear variance of the data set. There is also one popular method which is t- SNE (t-stochastic neighbour embedding). This method reduces non linear dimensions.
   
   Example: When we analyze one image, we get many pixels which are not necessary for our job. If we have to reduce the unnecessary pixels then we can use this method to do so.
   
5. Ensemble methods: In this method, we use different prediction models. It means that we assemble different models to increase our efficiency. Let’s understand it by an example. You want to buy one laptop but you are not finding all good features in a single laptop that is available in market. Now, you can buy all the good things from the market and then assemble them to make your laptop and it’s efficiency will be maximum. In the same way ensemble method works.
   
   Example: Random forest and XGBoost are the example of ensemble methods. In random forest algorithm, we assemble multiple decision trees trained from different datasets. So, it is sure that the efficiency of forest will exceeds the efficiency of decision trees.
   
6. Neural networks and deep learning: In linear and logistic regression, we see that it detects linear pattern from the data. Neural network is used to detect non linear pattern in the data. Deep learning techniques require a lot of data and computation power for best performance as this method is self-tuning many parameters within vast architectures.
   
7. Transfer learning: Transfer learning is a machine learning technique in which we can use a part of neural network to do a new work. We can do this by changing some layers of neural network. The main advantage of this method is it reduces the required resources. It reduces both your time and money.
   
   Example: Suppose you are working on a model which can detect image of different types of shirts but suddenly authority has informed you that they want pant. Now, you can use transfer learning to reduce your time and cost.
   
8. Reinforcement learning:  When you have very less amount of data or no data then you can use reinforcement learning. In this model, we use experience of the model to evaluate next input.
   
   Example: Suppose, one mouse is finding cheese in a maze. At first it will be difficult for the mouse but in next time he will learn from his previous experience.
   
9. Natural language processing: Actually natural language processing is not properly a machine learning technique but it prepares text for machine learning.
   
   Example: When we type something in chats, sometimes the word completed automatically. In some cases machines make conversations with human. This things work by natural language processing.
   
10. Word embedding: TFM and TFIDF are numerical representations of textdocuments that consider only frequency and weighted frequencies to represent text documents. In contrast, word embedding can capture the context of a word in a document. As with word context, embeddings can measure similarity between words, allowing us to perform arithmetic with words.